Lines Matching +full:cpu +full:- +full:centric
2 * Copyright © 2015-2016 Intel Corporation
40 * captured by DMA from the GPU, unsynchronized with and unrelated to the CPU.
44 * without special privileges. Access to system-wide metrics requires root
58 * might sample sets of tightly-coupled counters, depending on the
70 * interleaved with event-type specific members.
74 * with the CPU, using HW specific packing formats for counter sets. Sometimes
76 * would be acceptable to expose them to unprivileged applications - to hide
88 * into perf's currently cpu centric design.
96 * side-band OA data captured via MI_REPORT_PERF_COUNT commands; we're
102 * For posterity, in case we might re-visit trying to adapt core perf to be
106 * - The perf based OA PMU driver broke some significant design assumptions:
108 * Existing perf pmus are used for profiling work on a cpu and we were
110 * implications, the need to fake cpu-related data (such as user/kernel
112 * as a way to forward device-specific status records.
115 * involvement from the CPU, making our PMU driver the first of a kind.
117 * Given the way we were periodically forward data from the GPU-mapped, OA
123 * explicitly initiated from the cpu (say in response to a userspace read())
125 * trigger a report from the cpu on demand.
130 * opened, there's no clear precedent for being able to provide group-wide
139 * for combining with the side-band raw reports it captures using
142 * - As a side note on perf's grouping feature; there was also some concern
158 * one time. The OA unit is not designed to allow re-configuration while in
178 * - It felt like our perf based PMU was making some technical compromises
182 * cpu core, while our device pmu related to neither. Events opened with a
184 * that process - so not appropriate for us. When an event is related to a
185 * cpu id, perf ensures pmu methods will be invoked via an inter process
187 * perf events for a specific cpu. This was workable but it meant the
214 #define OA_TAKEN(tail, head) ((tail - head) & (OA_BUFFER_SIZE - 1))
222 * CPU).
225 * by checking for a zeroed report-id field in tail reports, we want to account
245 * non-periodic reports (such as on context switch) or the OA unit may be
299 * code assumes all reports have a power-of-two size and ~(size - 1) can
327 * struct perf_open_properties - for validated properties given to open a stream
342 * @poll_oa_period: The period in nanoseconds at which the CPU will check for OA
386 kfree(oa_config->flex_regs); in i915_oa_config_release()
387 kfree(oa_config->b_counter_regs); in i915_oa_config_release()
388 kfree(oa_config->mux_regs); in i915_oa_config_release()
399 oa_config = idr_find(&perf->metrics_idr, metrics_set); in i915_perf_get_oa_config()
409 i915_oa_config_put(oa_bo->oa_config); in free_oa_config_bo()
410 i915_vma_put(oa_bo->vma); in free_oa_config_bo()
416 struct intel_uncore *uncore = stream->uncore; in gen12_oa_hw_tail_read()
424 struct intel_uncore *uncore = stream->uncore; in gen8_oa_hw_tail_read()
431 struct intel_uncore *uncore = stream->uncore; in gen7_oa_hw_tail_read()
438 * oa_buffer_check_unlocked - check for data and update tail ptr state
446 * pointer having a race with respect to what data is visible to the CPU.
463 u32 gtt_offset = i915_ggtt_offset(stream->oa_buffer.vma); in oa_buffer_check_unlocked()
464 int report_size = stream->oa_buffer.format_size; in oa_buffer_check_unlocked()
474 spin_lock_irqsave(&stream->oa_buffer.ptr_lock, flags); in oa_buffer_check_unlocked()
476 hw_tail = stream->perf->ops.oa_hw_tail_read(stream); in oa_buffer_check_unlocked()
481 hw_tail &= ~(report_size - 1); in oa_buffer_check_unlocked()
485 if (hw_tail == stream->oa_buffer.aging_tail && in oa_buffer_check_unlocked()
486 (now - stream->oa_buffer.aging_timestamp) > OA_TAIL_MARGIN_NSEC) { in oa_buffer_check_unlocked()
491 stream->oa_buffer.tail = stream->oa_buffer.aging_tail; in oa_buffer_check_unlocked()
497 * anywhere between this head and stream->oa_buffer.tail. in oa_buffer_check_unlocked()
499 head = stream->oa_buffer.head - gtt_offset; in oa_buffer_check_unlocked()
500 aged_tail = stream->oa_buffer.tail - gtt_offset; in oa_buffer_check_unlocked()
502 hw_tail -= gtt_offset; in oa_buffer_check_unlocked()
517 u32 *report32 = (void *)(stream->oa_buffer.vaddr + tail); in oa_buffer_check_unlocked()
522 tail = (tail - report_size) & (OA_BUFFER_SIZE - 1); in oa_buffer_check_unlocked()
526 __ratelimit(&stream->perf->tail_pointer_race)) in oa_buffer_check_unlocked()
531 stream->oa_buffer.tail = gtt_offset + tail; in oa_buffer_check_unlocked()
532 stream->oa_buffer.aging_tail = gtt_offset + hw_tail; in oa_buffer_check_unlocked()
533 stream->oa_buffer.aging_timestamp = now; in oa_buffer_check_unlocked()
536 pollin = OA_TAKEN(stream->oa_buffer.tail - gtt_offset, in oa_buffer_check_unlocked()
537 stream->oa_buffer.head - gtt_offset) >= report_size; in oa_buffer_check_unlocked()
539 spin_unlock_irqrestore(&stream->oa_buffer.ptr_lock, flags); in oa_buffer_check_unlocked()
545 * append_oa_status - Appends a status record to a userspace read() buffer.
546 * @stream: An i915-perf stream opened for OA metrics
567 if ((count - *offset) < header.size) in append_oa_status()
568 return -ENOSPC; in append_oa_status()
571 return -EFAULT; in append_oa_status()
579 * append_oa_sample - Copies single OA report into userspace read() buffer.
580 * @stream: An i915-perf stream opened for OA metrics
587 * properties when opening a stream, tracked as `stream->sample_flags`. This
601 int report_size = stream->oa_buffer.format_size; in append_oa_sample()
603 u32 sample_flags = stream->sample_flags; in append_oa_sample()
607 header.size = stream->sample_size; in append_oa_sample()
609 if ((count - *offset) < header.size) in append_oa_sample()
610 return -ENOSPC; in append_oa_sample()
614 return -EFAULT; in append_oa_sample()
619 return -EFAULT; in append_oa_sample()
629 * @stream: An i915-perf stream opened for OA metrics
634 * Notably any error condition resulting in a short read (-%ENOSPC or
635 * -%EFAULT) will be returned even though one or more records may
642 * and back-to-front you're not alone, but this follows the
652 struct intel_uncore *uncore = stream->uncore; in gen8_append_oa_reports()
653 int report_size = stream->oa_buffer.format_size; in gen8_append_oa_reports()
654 u8 *oa_buf_base = stream->oa_buffer.vaddr; in gen8_append_oa_reports()
655 u32 gtt_offset = i915_ggtt_offset(stream->oa_buffer.vma); in gen8_append_oa_reports()
656 u32 mask = (OA_BUFFER_SIZE - 1); in gen8_append_oa_reports()
663 if (drm_WARN_ON(&uncore->i915->drm, !stream->enabled)) in gen8_append_oa_reports()
664 return -EIO; in gen8_append_oa_reports()
666 spin_lock_irqsave(&stream->oa_buffer.ptr_lock, flags); in gen8_append_oa_reports()
668 head = stream->oa_buffer.head; in gen8_append_oa_reports()
669 tail = stream->oa_buffer.tail; in gen8_append_oa_reports()
671 spin_unlock_irqrestore(&stream->oa_buffer.ptr_lock, flags); in gen8_append_oa_reports()
677 head -= gtt_offset; in gen8_append_oa_reports()
678 tail -= gtt_offset; in gen8_append_oa_reports()
687 if (drm_WARN_ONCE(&uncore->i915->drm, in gen8_append_oa_reports()
692 return -EIO; in gen8_append_oa_reports()
712 if (drm_WARN_ON(&uncore->i915->drm, in gen8_append_oa_reports()
713 (OA_BUFFER_SIZE - head) < report_size)) { in gen8_append_oa_reports()
714 drm_err(&uncore->i915->drm, in gen8_append_oa_reports()
715 "Spurious OA head ptr: non-integral report offset\n"); in gen8_append_oa_reports()
729 (IS_GEN(stream->perf->i915, 12) ? in gen8_append_oa_reports()
733 if (__ratelimit(&stream->perf->spurious_report_rs)) in gen8_append_oa_reports()
738 ctx_id = report32[2] & stream->specific_ctx_id_mask; in gen8_append_oa_reports()
742 * invalid to be sure we avoid false-positive, single-context in gen8_append_oa_reports()
748 if (!(report32[0] & stream->perf->gen8_valid_ctx_bit) && in gen8_append_oa_reports()
749 INTEL_GEN(stream->perf->i915) <= 11) in gen8_append_oa_reports()
755 * stop the counters from updating as system-wide / global in gen8_append_oa_reports()
759 * filtered on the cpu but it's not worth trying to in gen8_append_oa_reports()
763 * provide a side-band view of the real values. in gen8_append_oa_reports()
767 * needs be forwarded bookend context-switch reports so that it in gen8_append_oa_reports()
780 * switches since it's not-uncommon for periodic samples to in gen8_append_oa_reports()
783 if (!stream->perf->exclusive_stream->ctx || in gen8_append_oa_reports()
784 stream->specific_ctx_id == ctx_id || in gen8_append_oa_reports()
785 stream->oa_buffer.last_ctx_id == stream->specific_ctx_id || in gen8_append_oa_reports()
792 if (stream->perf->exclusive_stream->ctx && in gen8_append_oa_reports()
793 stream->specific_ctx_id != ctx_id) { in gen8_append_oa_reports()
802 stream->oa_buffer.last_ctx_id = ctx_id; in gen8_append_oa_reports()
816 oaheadptr = IS_GEN(stream->perf->i915, 12) ? in gen8_append_oa_reports()
819 spin_lock_irqsave(&stream->oa_buffer.ptr_lock, flags); in gen8_append_oa_reports()
828 stream->oa_buffer.head = head; in gen8_append_oa_reports()
830 spin_unlock_irqrestore(&stream->oa_buffer.ptr_lock, flags); in gen8_append_oa_reports()
837 * gen8_oa_read - copy status records then buffered OA reports
838 * @stream: An i915-perf stream opened for OA metrics
861 struct intel_uncore *uncore = stream->uncore; in gen8_oa_read()
866 if (drm_WARN_ON(&uncore->i915->drm, !stream->oa_buffer.vaddr)) in gen8_oa_read()
867 return -EIO; in gen8_oa_read()
869 oastatus_reg = IS_GEN(stream->perf->i915, 12) ? in gen8_oa_read()
895 stream->period_exponent); in gen8_oa_read()
897 stream->perf->ops.oa_disable(stream); in gen8_oa_read()
898 stream->perf->ops.oa_enable(stream); in gen8_oa_read()
901 * Note: .oa_enable() is expected to re-init the oabuffer and in gen8_oa_read()
916 IS_GEN_RANGE(uncore->i915, 8, 10) ? in gen8_oa_read()
926 * @stream: An i915-perf stream opened for OA metrics
931 * Notably any error condition resulting in a short read (-%ENOSPC or
932 * -%EFAULT) will be returned even though one or more records may
939 * and back-to-front you're not alone, but this follows the
949 struct intel_uncore *uncore = stream->uncore; in gen7_append_oa_reports()
950 int report_size = stream->oa_buffer.format_size; in gen7_append_oa_reports()
951 u8 *oa_buf_base = stream->oa_buffer.vaddr; in gen7_append_oa_reports()
952 u32 gtt_offset = i915_ggtt_offset(stream->oa_buffer.vma); in gen7_append_oa_reports()
953 u32 mask = (OA_BUFFER_SIZE - 1); in gen7_append_oa_reports()
960 if (drm_WARN_ON(&uncore->i915->drm, !stream->enabled)) in gen7_append_oa_reports()
961 return -EIO; in gen7_append_oa_reports()
963 spin_lock_irqsave(&stream->oa_buffer.ptr_lock, flags); in gen7_append_oa_reports()
965 head = stream->oa_buffer.head; in gen7_append_oa_reports()
966 tail = stream->oa_buffer.tail; in gen7_append_oa_reports()
968 spin_unlock_irqrestore(&stream->oa_buffer.ptr_lock, flags); in gen7_append_oa_reports()
973 head -= gtt_offset; in gen7_append_oa_reports()
974 tail -= gtt_offset; in gen7_append_oa_reports()
982 if (drm_WARN_ONCE(&uncore->i915->drm, in gen7_append_oa_reports()
987 return -EIO; in gen7_append_oa_reports()
1004 if (drm_WARN_ON(&uncore->i915->drm, in gen7_append_oa_reports()
1005 (OA_BUFFER_SIZE - head) < report_size)) { in gen7_append_oa_reports()
1006 drm_err(&uncore->i915->drm, in gen7_append_oa_reports()
1007 "Spurious OA head ptr: non-integral report offset\n"); in gen7_append_oa_reports()
1011 /* The report-ID field for periodic samples includes in gen7_append_oa_reports()
1018 if (__ratelimit(&stream->perf->spurious_report_rs)) in gen7_append_oa_reports()
1035 spin_lock_irqsave(&stream->oa_buffer.ptr_lock, flags); in gen7_append_oa_reports()
1045 stream->oa_buffer.head = head; in gen7_append_oa_reports()
1047 spin_unlock_irqrestore(&stream->oa_buffer.ptr_lock, flags); in gen7_append_oa_reports()
1054 * gen7_oa_read - copy status records then buffered OA reports
1055 * @stream: An i915-perf stream opened for OA metrics
1074 struct intel_uncore *uncore = stream->uncore; in gen7_oa_read()
1078 if (drm_WARN_ON(&uncore->i915->drm, !stream->oa_buffer.vaddr)) in gen7_oa_read()
1079 return -EIO; in gen7_oa_read()
1088 oastatus1 &= ~stream->perf->gen7_latched_oastatus1; in gen7_oa_read()
1092 * - The status can be interpreted to mean that the buffer is in gen7_oa_read()
1094 * which will start to report a near-empty buffer after an in gen7_oa_read()
1099 * - Since it also implies the HW has started overwriting old in gen7_oa_read()
1104 * - In the future we may want to introduce a flight recorder in gen7_oa_read()
1117 stream->period_exponent); in gen7_oa_read()
1119 stream->perf->ops.oa_disable(stream); in gen7_oa_read()
1120 stream->perf->ops.oa_enable(stream); in gen7_oa_read()
1130 stream->perf->gen7_latched_oastatus1 |= in gen7_oa_read()
1138 * i915_oa_wait_unlocked - handles blocking IO until OA data available
1139 * @stream: An i915-perf stream opened for OA metrics
1142 * for OA metrics. It waits until the hrtimer callback finds a non-empty
1146 * since any subsequent read handling will return -EAGAIN if there isn't
1154 if (!stream->periodic) in i915_oa_wait_unlocked()
1155 return -EIO; in i915_oa_wait_unlocked()
1157 return wait_event_interruptible(stream->poll_wq, in i915_oa_wait_unlocked()
1162 * i915_oa_poll_wait - call poll_wait() for an OA stream poll()
1163 * @stream: An i915-perf stream opened for OA metrics
1175 poll_wait(file, &stream->poll_wq, wait); in i915_oa_poll_wait()
1179 * i915_oa_read - just calls through to &i915_oa_ops->read
1180 * @stream: An i915-perf stream opened for OA metrics
1195 return stream->perf->ops.read(stream, buf, count, offset); in i915_oa_read()
1201 struct i915_gem_context *ctx = stream->ctx; in oa_pin_context()
1204 int err = -ENODEV; in oa_pin_context()
1207 if (ce->engine != stream->engine) /* first match! */ in oa_pin_context()
1225 if (err == -EDEADLK) { in oa_pin_context()
1235 stream->pinned_ctx = ce; in oa_pin_context()
1236 return stream->pinned_ctx; in oa_pin_context()
1240 * oa_get_render_ctx_id - determine and hold ctx hw id
1241 * @stream: An i915-perf stream opened for OA metrics
1257 switch (INTEL_GEN(ce->engine->i915)) { in oa_get_render_ctx_id()
1263 stream->specific_ctx_id = i915_ggtt_offset(ce->state); in oa_get_render_ctx_id()
1264 stream->specific_ctx_id_mask = 0; in oa_get_render_ctx_id()
1271 if (intel_engine_in_execlists_submission_mode(ce->engine)) { in oa_get_render_ctx_id()
1272 stream->specific_ctx_id_mask = in oa_get_render_ctx_id()
1273 (1U << GEN8_CTX_ID_WIDTH) - 1; in oa_get_render_ctx_id()
1274 stream->specific_ctx_id = stream->specific_ctx_id_mask; in oa_get_render_ctx_id()
1286 stream->specific_ctx_id = ce->lrc.lrca >> 12; in oa_get_render_ctx_id()
1292 stream->specific_ctx_id_mask = in oa_get_render_ctx_id()
1293 (1U << (GEN8_CTX_ID_WIDTH - 1)) - 1; in oa_get_render_ctx_id()
1299 stream->specific_ctx_id_mask = in oa_get_render_ctx_id()
1300 ((1U << GEN11_SW_CTX_ID_WIDTH) - 1) << (GEN11_SW_CTX_ID_SHIFT - 32); in oa_get_render_ctx_id()
1303 * 0 - BITS_PER_LONG are used by other contexts in oa_get_render_ctx_id()
1306 stream->specific_ctx_id = (GEN12_MAX_CONTEXT_HW_ID - 1) << (GEN11_SW_CTX_ID_SHIFT - 32); in oa_get_render_ctx_id()
1311 MISSING_CASE(INTEL_GEN(ce->engine->i915)); in oa_get_render_ctx_id()
1314 ce->tag = stream->specific_ctx_id; in oa_get_render_ctx_id()
1316 drm_dbg(&stream->perf->i915->drm, in oa_get_render_ctx_id()
1318 stream->specific_ctx_id, in oa_get_render_ctx_id()
1319 stream->specific_ctx_id_mask); in oa_get_render_ctx_id()
1325 * oa_put_render_ctx_id - counterpart to oa_get_render_ctx_id releases hold
1326 * @stream: An i915-perf stream opened for OA metrics
1335 ce = fetch_and_zero(&stream->pinned_ctx); in oa_put_render_ctx_id()
1337 ce->tag = 0; /* recomputed on next submission after parking */ in oa_put_render_ctx_id()
1341 stream->specific_ctx_id = INVALID_CTX_ID; in oa_put_render_ctx_id()
1342 stream->specific_ctx_id_mask = 0; in oa_put_render_ctx_id()
1348 i915_vma_unpin_and_release(&stream->oa_buffer.vma, in free_oa_buffer()
1351 stream->oa_buffer.vaddr = NULL; in free_oa_buffer()
1359 i915_oa_config_put(stream->oa_config); in free_oa_configs()
1360 llist_for_each_entry_safe(oa_bo, tmp, stream->oa_config_bos.first, node) in free_oa_configs()
1367 i915_vma_unpin_and_release(&stream->noa_wait, 0); in free_noa_wait()
1372 struct i915_perf *perf = stream->perf; in i915_oa_stream_destroy()
1374 BUG_ON(stream != perf->exclusive_stream); in i915_oa_stream_destroy()
1382 WRITE_ONCE(perf->exclusive_stream, NULL); in i915_oa_stream_destroy()
1383 perf->ops.disable_metric_set(stream); in i915_oa_stream_destroy()
1387 intel_uncore_forcewake_put(stream->uncore, FORCEWAKE_ALL); in i915_oa_stream_destroy()
1388 intel_engine_pm_put(stream->engine); in i915_oa_stream_destroy()
1390 if (stream->ctx) in i915_oa_stream_destroy()
1396 if (perf->spurious_report_rs.missed) { in i915_oa_stream_destroy()
1398 perf->spurious_report_rs.missed); in i915_oa_stream_destroy()
1404 struct intel_uncore *uncore = stream->uncore; in gen7_init_oa_buffer()
1405 u32 gtt_offset = i915_ggtt_offset(stream->oa_buffer.vma); in gen7_init_oa_buffer()
1408 spin_lock_irqsave(&stream->oa_buffer.ptr_lock, flags); in gen7_init_oa_buffer()
1410 /* Pre-DevBDW: OABUFFER must be set with counters off, in gen7_init_oa_buffer()
1415 stream->oa_buffer.head = gtt_offset; in gen7_init_oa_buffer()
1423 stream->oa_buffer.aging_tail = INVALID_TAIL_PTR; in gen7_init_oa_buffer()
1424 stream->oa_buffer.tail = gtt_offset; in gen7_init_oa_buffer()
1426 spin_unlock_irqrestore(&stream->oa_buffer.ptr_lock, flags); in gen7_init_oa_buffer()
1432 stream->perf->gen7_latched_oastatus1 = 0; in gen7_init_oa_buffer()
1436 * first allocating), we may re-init the OA buffer, either in gen7_init_oa_buffer()
1437 * when re-enabling a stream or in error/reset paths. in gen7_init_oa_buffer()
1439 * The reason we clear the buffer for each re-init is for the in gen7_init_oa_buffer()
1441 * report-id field to make sure it's non-zero which relies on in gen7_init_oa_buffer()
1445 memset(stream->oa_buffer.vaddr, 0, OA_BUFFER_SIZE); in gen7_init_oa_buffer()
1450 struct intel_uncore *uncore = stream->uncore; in gen8_init_oa_buffer()
1451 u32 gtt_offset = i915_ggtt_offset(stream->oa_buffer.vma); in gen8_init_oa_buffer()
1454 spin_lock_irqsave(&stream->oa_buffer.ptr_lock, flags); in gen8_init_oa_buffer()
1458 stream->oa_buffer.head = gtt_offset; in gen8_init_oa_buffer()
1475 stream->oa_buffer.aging_tail = INVALID_TAIL_PTR; in gen8_init_oa_buffer()
1476 stream->oa_buffer.tail = gtt_offset; in gen8_init_oa_buffer()
1483 stream->oa_buffer.last_ctx_id = INVALID_CTX_ID; in gen8_init_oa_buffer()
1485 spin_unlock_irqrestore(&stream->oa_buffer.ptr_lock, flags); in gen8_init_oa_buffer()
1490 * first allocating), we may re-init the OA buffer, either in gen8_init_oa_buffer()
1491 * when re-enabling a stream or in error/reset paths. in gen8_init_oa_buffer()
1493 * The reason we clear the buffer for each re-init is for the in gen8_init_oa_buffer()
1495 * reason field to make sure it's non-zero which relies on in gen8_init_oa_buffer()
1499 memset(stream->oa_buffer.vaddr, 0, OA_BUFFER_SIZE); in gen8_init_oa_buffer()
1504 struct intel_uncore *uncore = stream->uncore; in gen12_init_oa_buffer()
1505 u32 gtt_offset = i915_ggtt_offset(stream->oa_buffer.vma); in gen12_init_oa_buffer()
1508 spin_lock_irqsave(&stream->oa_buffer.ptr_lock, flags); in gen12_init_oa_buffer()
1513 stream->oa_buffer.head = gtt_offset; in gen12_init_oa_buffer()
1529 stream->oa_buffer.aging_tail = INVALID_TAIL_PTR; in gen12_init_oa_buffer()
1530 stream->oa_buffer.tail = gtt_offset; in gen12_init_oa_buffer()
1537 stream->oa_buffer.last_ctx_id = INVALID_CTX_ID; in gen12_init_oa_buffer()
1539 spin_unlock_irqrestore(&stream->oa_buffer.ptr_lock, flags); in gen12_init_oa_buffer()
1544 * first allocating), we may re-init the OA buffer, either in gen12_init_oa_buffer()
1545 * when re-enabling a stream or in error/reset paths. in gen12_init_oa_buffer()
1547 * The reason we clear the buffer for each re-init is for the in gen12_init_oa_buffer()
1549 * reason field to make sure it's non-zero which relies on in gen12_init_oa_buffer()
1553 memset(stream->oa_buffer.vaddr, 0, in gen12_init_oa_buffer()
1554 stream->oa_buffer.vma->size); in gen12_init_oa_buffer()
1559 struct drm_i915_private *i915 = stream->perf->i915; in alloc_oa_buffer()
1564 if (drm_WARN_ON(&i915->drm, stream->oa_buffer.vma)) in alloc_oa_buffer()
1565 return -ENODEV; in alloc_oa_buffer()
1570 bo = i915_gem_object_create_shmem(stream->perf->i915, OA_BUFFER_SIZE); in alloc_oa_buffer()
1572 drm_err(&i915->drm, "Failed to allocate OA buffer\n"); in alloc_oa_buffer()
1584 stream->oa_buffer.vma = vma; in alloc_oa_buffer()
1586 stream->oa_buffer.vaddr = in alloc_oa_buffer()
1588 if (IS_ERR(stream->oa_buffer.vaddr)) { in alloc_oa_buffer()
1589 ret = PTR_ERR(stream->oa_buffer.vaddr); in alloc_oa_buffer()
1601 stream->oa_buffer.vaddr = NULL; in alloc_oa_buffer()
1602 stream->oa_buffer.vma = NULL; in alloc_oa_buffer()
1616 if (INTEL_GEN(stream->perf->i915) >= 8) in save_restore_register()
1622 *cs++ = intel_gt_scratch_offset(stream->engine->gt, in save_restore_register()
1632 struct drm_i915_private *i915 = stream->perf->i915; in alloc_noa_wait()
1635 const u64 delay_ticks = 0xffffffffffffffff - in alloc_noa_wait()
1636 i915_cs_timestamp_ns_to_ticks(i915, atomic64_read(&stream->perf->noa_programming_delay)); in alloc_noa_wait()
1637 const u32 base = stream->engine->mmio_base; in alloc_noa_wait()
1652 drm_err(&i915->drm, in alloc_noa_wait()
1694 *cs++ = MI_LOAD_REGISTER_REG | (3 - 2); in alloc_noa_wait()
1712 *cs++ = MI_LOAD_REGISTER_REG | (3 - 2); in alloc_noa_wait()
1732 *cs++ = MI_LOAD_REGISTER_REG | (3 - 2); in alloc_noa_wait()
1741 *cs++ = i915_ggtt_offset(vma) + (ts0 - batch) * 4; in alloc_noa_wait()
1746 * (((1 * << 64) - 1) - delay_ns) in alloc_noa_wait()
1769 *cs++ = MI_LOAD_REGISTER_REG | (3 - 2); in alloc_noa_wait()
1778 *cs++ = i915_ggtt_offset(vma) + (jump - batch) * 4; in alloc_noa_wait()
1793 GEM_BUG_ON(cs - batch > PAGE_SIZE / sizeof(*batch)); in alloc_noa_wait()
1798 stream->noa_wait = vma; in alloc_noa_wait()
1817 n_regs - i, in write_cs_mi_lri()
1853 return ERR_PTR(-ENOMEM); in alloc_oa_config_buffer()
1855 config_length += num_lri_dwords(oa_config->mux_regs_len); in alloc_oa_config_buffer()
1856 config_length += num_lri_dwords(oa_config->b_counter_regs_len); in alloc_oa_config_buffer()
1857 config_length += num_lri_dwords(oa_config->flex_regs_len); in alloc_oa_config_buffer()
1861 obj = i915_gem_object_create_shmem(stream->perf->i915, config_length); in alloc_oa_config_buffer()
1874 oa_config->mux_regs, in alloc_oa_config_buffer()
1875 oa_config->mux_regs_len); in alloc_oa_config_buffer()
1877 oa_config->b_counter_regs, in alloc_oa_config_buffer()
1878 oa_config->b_counter_regs_len); in alloc_oa_config_buffer()
1880 oa_config->flex_regs, in alloc_oa_config_buffer()
1881 oa_config->flex_regs_len); in alloc_oa_config_buffer()
1884 *cs++ = (INTEL_GEN(stream->perf->i915) < 8 ? in alloc_oa_config_buffer()
1887 *cs++ = i915_ggtt_offset(stream->noa_wait); in alloc_oa_config_buffer()
1893 oa_bo->vma = i915_vma_instance(obj, in alloc_oa_config_buffer()
1894 &stream->engine->gt->ggtt->vm, in alloc_oa_config_buffer()
1896 if (IS_ERR(oa_bo->vma)) { in alloc_oa_config_buffer()
1897 err = PTR_ERR(oa_bo->vma); in alloc_oa_config_buffer()
1901 oa_bo->oa_config = i915_oa_config_get(oa_config); in alloc_oa_config_buffer()
1902 llist_add(&oa_bo->node, &stream->oa_config_bos); in alloc_oa_config_buffer()
1922 llist_for_each_entry(oa_bo, stream->oa_config_bos.first, node) { in get_oa_vma()
1923 if (oa_bo->oa_config == oa_config && in get_oa_vma()
1924 memcmp(oa_bo->oa_config->uuid, in get_oa_vma()
1925 oa_config->uuid, in get_oa_vma()
1926 sizeof(oa_config->uuid)) == 0) in get_oa_vma()
1935 return i915_vma_get(oa_bo->vma); in get_oa_vma()
1955 err = i915_gem_object_lock(vma->obj, &ww); in emit_oa_config()
1963 intel_engine_pm_get(ce->engine); in emit_oa_config()
1965 intel_engine_pm_put(ce->engine); in emit_oa_config()
1983 err = i915_request_await_object(rq, vma->obj, 0); in emit_oa_config()
1989 err = rq->engine->emit_bb_start(rq, in emit_oa_config()
1990 vma->node.start, 0, in emit_oa_config()
2000 if (err == -EDEADLK) { in emit_oa_config()
2013 return stream->pinned_ctx ?: stream->engine->kernel_context; in oa_context()
2020 struct intel_uncore *uncore = stream->uncore; in hsw_enable_metric_set()
2027 * unable to count the events from non-render clock domain. in hsw_enable_metric_set()
2029 * count the events from non-render domain. Unit level clock in hsw_enable_metric_set()
2038 stream->oa_config, oa_context(stream), in hsw_enable_metric_set()
2044 struct intel_uncore *uncore = stream->uncore; in hsw_disable_metric_set()
2068 for (i = 0; i < oa_config->flex_regs_len; i++) { in oa_config_flex_reg()
2069 if (i915_mmio_reg_offset(oa_config->flex_regs[i].addr) == mmio) in oa_config_flex_reg()
2070 return oa_config->flex_regs[i].value; in oa_config_flex_reg()
2079 * It's fine to put out-of-date values into these per-context registers
2086 u32 ctx_oactxctrl = stream->perf->ctx_oactxctrl_offset; in gen8_update_reg_state_unlocked()
2087 u32 ctx_flexeu0 = stream->perf->ctx_flexeu0_offset; in gen8_update_reg_state_unlocked()
2098 u32 *reg_state = ce->lrc_reg_state; in gen8_update_reg_state_unlocked()
2102 (stream->period_exponent << GEN8_OA_TIMER_PERIOD_SHIFT) | in gen8_update_reg_state_unlocked()
2103 (stream->periodic ? GEN8_OA_TIMER_ENABLE : 0) | in gen8_update_reg_state_unlocked()
2108 oa_config_flex_reg(stream->oa_config, flex_regs[i]); in gen8_update_reg_state_unlocked()
2129 offset = i915_ggtt_offset(ce->state) + LRC_STATE_OFFSET; in gen8_store_flex()
2132 *cs++ = offset + flex->offset * sizeof(u32); in gen8_store_flex()
2134 *cs++ = flex->value; in gen8_store_flex()
2135 } while (flex++, --count); in gen8_store_flex()
2157 *cs++ = i915_mmio_reg_offset(flex->reg); in gen8_load_flex()
2158 *cs++ = flex->value; in gen8_load_flex()
2159 } while (flex++, --count); in gen8_load_flex()
2173 rq = intel_engine_create_kernel_request(ce->engine); in gen8_modify_context()
2194 intel_engine_pm_get(ce->engine); in gen8_modify_self()
2196 intel_engine_pm_put(ce->engine); in gen8_modify_self()
2223 GEM_BUG_ON(ce == ce->engine->kernel_context); in gen8_configure_context()
2225 if (ce->engine->class != RENDER_CLASS) in gen8_configure_context()
2232 flex->value = intel_sseu_make_rpcs(ce->engine->gt, &ce->sseu); in gen8_configure_context()
2248 struct intel_context *ce = stream->pinned_ctx; in gen12_configure_oar_context()
2249 u32 format = stream->oa_buffer.format; in gen12_configure_oar_context()
2253 stream->perf->ctx_oactxctrl_offset + 1, in gen12_configure_oar_context()
2269 RING_CONTEXT_CONTROL(ce->engine->mmio_base), in gen12_configure_oar_context()
2293 * Manages updating the per-context aspects of the OA stream
2303 * won't automatically reload an out-of-date timer exponent even
2307 * - Ensure the currently running context's per-context OA state is
2309 * - Ensure that all existing contexts will have the correct per-context
2311 * - Ensure any new contexts will be initialized with the correct
2312 * per-context OA state.
2323 struct drm_i915_private *i915 = stream->perf->i915; in oa_configure_all_contexts()
2328 lockdep_assert_held(&stream->perf->lock); in oa_configure_all_contexts()
2333 * lite-restore). This means we can't safely update a context's image, in oa_configure_all_contexts()
2346 spin_lock(&i915->gem.contexts.lock); in oa_configure_all_contexts()
2347 list_for_each_entry_safe(ctx, cn, &i915->gem.contexts.list, link) { in oa_configure_all_contexts()
2348 if (!kref_get_unless_zero(&ctx->ref)) in oa_configure_all_contexts()
2351 spin_unlock(&i915->gem.contexts.lock); in oa_configure_all_contexts()
2359 spin_lock(&i915->gem.contexts.lock); in oa_configure_all_contexts()
2363 spin_unlock(&i915->gem.contexts.lock); in oa_configure_all_contexts()
2371 struct intel_context *ce = engine->kernel_context; in oa_configure_all_contexts()
2373 if (engine->class != RENDER_CLASS) in oa_configure_all_contexts()
2376 regs[0].value = intel_sseu_make_rpcs(engine->gt, &ce->sseu); in oa_configure_all_contexts()
2409 const u32 ctx_flexeu0 = stream->perf->ctx_flexeu0_offset; in lrc_configure_all_contexts()
2418 stream->perf->ctx_oactxctrl_offset + 1, in lrc_configure_all_contexts()
2432 (stream->period_exponent << GEN8_OA_TIMER_PERIOD_SHIFT) | in lrc_configure_all_contexts()
2433 (stream->periodic ? GEN8_OA_TIMER_ENABLE : 0) | in lrc_configure_all_contexts()
2448 struct intel_uncore *uncore = stream->uncore; in gen8_enable_metric_set()
2449 struct i915_oa_config *oa_config = stream->oa_config; in gen8_enable_metric_set()
2464 * Currently none of the high-level metrics we have depend on knowing in gen8_enable_metric_set()
2475 if (IS_GEN_RANGE(stream->perf->i915, 9, 11)) { in gen8_enable_metric_set()
2491 stream->oa_config, oa_context(stream), in gen8_enable_metric_set()
2498 (stream->sample_flags & SAMPLE_OA_REPORT) ? in oag_report_ctx_switches()
2506 struct intel_uncore *uncore = stream->uncore; in gen12_enable_metric_set()
2507 struct i915_oa_config *oa_config = stream->oa_config; in gen12_enable_metric_set()
2508 bool periodic = stream->periodic; in gen12_enable_metric_set()
2509 u32 period_exponent = stream->period_exponent; in gen12_enable_metric_set()
2542 if (stream->ctx) { in gen12_enable_metric_set()
2549 stream->oa_config, oa_context(stream), in gen12_enable_metric_set()
2555 struct intel_uncore *uncore = stream->uncore; in gen8_disable_metric_set()
2565 struct intel_uncore *uncore = stream->uncore; in gen10_disable_metric_set()
2576 struct intel_uncore *uncore = stream->uncore; in gen12_disable_metric_set()
2582 if (stream->ctx) in gen12_disable_metric_set()
2591 struct intel_uncore *uncore = stream->uncore; in gen7_oa_enable()
2592 struct i915_gem_context *ctx = stream->ctx; in gen7_oa_enable()
2593 u32 ctx_id = stream->specific_ctx_id; in gen7_oa_enable()
2594 bool periodic = stream->periodic; in gen7_oa_enable()
2595 u32 period_exponent = stream->period_exponent; in gen7_oa_enable()
2596 u32 report_format = stream->oa_buffer.format; in gen7_oa_enable()
2621 struct intel_uncore *uncore = stream->uncore; in gen8_oa_enable()
2622 u32 report_format = stream->oa_buffer.format; in gen8_oa_enable()
2637 * filtering and instead filter on the cpu based on the context-id in gen8_oa_enable()
2647 struct intel_uncore *uncore = stream->uncore; in gen12_oa_enable()
2648 u32 report_format = stream->oa_buffer.format; in gen12_oa_enable()
2654 if (!(stream->sample_flags & SAMPLE_OA_REPORT)) in gen12_oa_enable()
2665 * i915_oa_stream_enable - handle `I915_PERF_IOCTL_ENABLE` for OA stream
2675 stream->pollin = false; in i915_oa_stream_enable()
2677 stream->perf->ops.oa_enable(stream); in i915_oa_stream_enable()
2679 if (stream->periodic) in i915_oa_stream_enable()
2680 hrtimer_start(&stream->poll_check_timer, in i915_oa_stream_enable()
2681 ns_to_ktime(stream->poll_oa_period), in i915_oa_stream_enable()
2687 struct intel_uncore *uncore = stream->uncore; in gen7_oa_disable()
2693 drm_err(&stream->perf->i915->drm, in gen7_oa_disable()
2699 struct intel_uncore *uncore = stream->uncore; in gen8_oa_disable()
2705 drm_err(&stream->perf->i915->drm, in gen8_oa_disable()
2711 struct intel_uncore *uncore = stream->uncore; in gen12_oa_disable()
2718 drm_err(&stream->perf->i915->drm, in gen12_oa_disable()
2726 drm_err(&stream->perf->i915->drm, in gen12_oa_disable()
2731 * i915_oa_stream_disable - handle `I915_PERF_IOCTL_DISABLE` for OA stream
2740 stream->perf->ops.oa_disable(stream); in i915_oa_stream_disable()
2742 if (stream->periodic) in i915_oa_stream_disable()
2743 hrtimer_cancel(&stream->poll_check_timer); in i915_oa_stream_disable()
2762 return -ENOMEM; in i915_perf_stream_enable_sync()
2764 err = stream->perf->ops.enable_metric_set(stream, active); in i915_perf_stream_enable_sync()
2776 const struct sseu_dev_info *devinfo_sseu = &engine->gt->info.sseu; in get_default_sseu_config()
2780 if (IS_GEN(engine->i915, 11)) { in get_default_sseu_config()
2783 * we select - just turn off low bits in the amount of half of in get_default_sseu_config()
2786 out_sseu->subslice_mask = in get_default_sseu_config()
2787 ~(~0 << (hweight8(out_sseu->subslice_mask) / 2)); in get_default_sseu_config()
2788 out_sseu->slice_mask = 0x1; in get_default_sseu_config()
2797 if (drm_sseu->engine.engine_class != engine->uabi_class || in get_sseu_config()
2798 drm_sseu->engine.engine_instance != engine->uabi_instance) in get_sseu_config()
2799 return -EINVAL; in get_sseu_config()
2801 return i915_gem_user_to_context_sseu(engine->gt, drm_sseu, out_sseu); in get_sseu_config()
2805 * i915_oa_stream_init - validate combined props for OA stream and init
2826 struct drm_i915_private *i915 = stream->perf->i915; in i915_oa_stream_init()
2827 struct i915_perf *perf = stream->perf; in i915_oa_stream_init()
2831 if (!props->engine) { in i915_oa_stream_init()
2833 return -EINVAL; in i915_oa_stream_init()
2841 if (!perf->metrics_kobj) { in i915_oa_stream_init()
2843 return -EINVAL; in i915_oa_stream_init()
2846 if (!(props->sample_flags & SAMPLE_OA_REPORT) && in i915_oa_stream_init()
2847 (INTEL_GEN(perf->i915) < 12 || !stream->ctx)) { in i915_oa_stream_init()
2849 return -EINVAL; in i915_oa_stream_init()
2852 if (!perf->ops.enable_metric_set) { in i915_oa_stream_init()
2854 return -ENODEV; in i915_oa_stream_init()
2862 if (perf->exclusive_stream) { in i915_oa_stream_init()
2864 return -EBUSY; in i915_oa_stream_init()
2867 if (!props->oa_format) { in i915_oa_stream_init()
2869 return -EINVAL; in i915_oa_stream_init()
2872 stream->engine = props->engine; in i915_oa_stream_init()
2873 stream->uncore = stream->engine->gt->uncore; in i915_oa_stream_init()
2875 stream->sample_size = sizeof(struct drm_i915_perf_record_header); in i915_oa_stream_init()
2877 format_size = perf->oa_formats[props->oa_format].size; in i915_oa_stream_init()
2879 stream->sample_flags = props->sample_flags; in i915_oa_stream_init()
2880 stream->sample_size += format_size; in i915_oa_stream_init()
2882 stream->oa_buffer.format_size = format_size; in i915_oa_stream_init()
2883 if (drm_WARN_ON(&i915->drm, stream->oa_buffer.format_size == 0)) in i915_oa_stream_init()
2884 return -EINVAL; in i915_oa_stream_init()
2886 stream->hold_preemption = props->hold_preemption; in i915_oa_stream_init()
2888 stream->oa_buffer.format = in i915_oa_stream_init()
2889 perf->oa_formats[props->oa_format].format; in i915_oa_stream_init()
2891 stream->periodic = props->oa_periodic; in i915_oa_stream_init()
2892 if (stream->periodic) in i915_oa_stream_init()
2893 stream->period_exponent = props->oa_period_exponent; in i915_oa_stream_init()
2895 if (stream->ctx) { in i915_oa_stream_init()
2909 stream->oa_config = i915_perf_get_oa_config(perf, props->metrics_set); in i915_oa_stream_init()
2910 if (!stream->oa_config) { in i915_oa_stream_init()
2911 DRM_DEBUG("Invalid OA config id=%i\n", props->metrics_set); in i915_oa_stream_init()
2912 ret = -EINVAL; in i915_oa_stream_init()
2916 /* PRM - observability performance counters: in i915_oa_stream_init()
2928 intel_engine_pm_get(stream->engine); in i915_oa_stream_init()
2929 intel_uncore_forcewake_get(stream->uncore, FORCEWAKE_ALL); in i915_oa_stream_init()
2935 stream->ops = &i915_oa_stream_ops; in i915_oa_stream_init()
2937 perf->sseu = props->sseu; in i915_oa_stream_init()
2938 WRITE_ONCE(perf->exclusive_stream, stream); in i915_oa_stream_init()
2947 stream->oa_config->uuid); in i915_oa_stream_init()
2949 hrtimer_init(&stream->poll_check_timer, in i915_oa_stream_init()
2951 stream->poll_check_timer.function = oa_poll_check_timer_cb; in i915_oa_stream_init()
2952 init_waitqueue_head(&stream->poll_wq); in i915_oa_stream_init()
2953 spin_lock_init(&stream->oa_buffer.ptr_lock); in i915_oa_stream_init()
2958 WRITE_ONCE(perf->exclusive_stream, NULL); in i915_oa_stream_init()
2959 perf->ops.disable_metric_set(stream); in i915_oa_stream_init()
2966 intel_uncore_forcewake_put(stream->uncore, FORCEWAKE_ALL); in i915_oa_stream_init()
2967 intel_engine_pm_put(stream->engine); in i915_oa_stream_init()
2973 if (stream->ctx) in i915_oa_stream_init()
2984 if (engine->class != RENDER_CLASS) in i915_oa_init_reg_state()
2988 stream = READ_ONCE(engine->i915->perf.exclusive_stream); in i915_oa_init_reg_state()
2989 if (stream && INTEL_GEN(stream->perf->i915) < 12) in i915_oa_init_reg_state()
2994 * i915_perf_read - handles read() FOP for i915 perf stream FDs
3002 * &i915_perf_stream_ops->read but to save having stream implementations (of
3016 struct i915_perf_stream *stream = file->private_data; in i915_perf_read()
3017 struct i915_perf *perf = stream->perf; in i915_perf_read()
3025 if (!stream->enabled) in i915_perf_read()
3026 return -EIO; in i915_perf_read()
3028 if (!(file->f_flags & O_NONBLOCK)) { in i915_perf_read()
3030 * stream->ops->wait_unlocked. in i915_perf_read()
3037 ret = stream->ops->wait_unlocked(stream); in i915_perf_read()
3041 mutex_lock(&perf->lock); in i915_perf_read()
3042 ret = stream->ops->read(stream, buf, count, &offset); in i915_perf_read()
3043 mutex_unlock(&perf->lock); in i915_perf_read()
3046 mutex_lock(&perf->lock); in i915_perf_read()
3047 ret = stream->ops->read(stream, buf, count, &offset); in i915_perf_read()
3048 mutex_unlock(&perf->lock); in i915_perf_read()
3055 * and read() returning -EAGAIN. Clearing the oa.pollin state here in i915_perf_read()
3058 * The exception to this is if ops->read() returned -ENOSPC which means in i915_perf_read()
3062 if (ret != -ENOSPC) in i915_perf_read()
3063 stream->pollin = false; in i915_perf_read()
3065 /* Possible values for ret are 0, -EFAULT, -ENOSPC, -EIO, ... */ in i915_perf_read()
3066 return offset ?: (ret ?: -EAGAIN); in i915_perf_read()
3075 stream->pollin = true; in oa_poll_check_timer_cb()
3076 wake_up(&stream->poll_wq); in oa_poll_check_timer_cb()
3080 ns_to_ktime(stream->poll_oa_period)); in oa_poll_check_timer_cb()
3086 * i915_perf_poll_locked - poll_wait() with a suitable wait queue for stream
3092 * &i915_perf_stream_ops->poll_wait to call poll_wait() with a wait queue that
3095 * Note: The &perf->lock mutex has been taken to serialize
3096 * with any non-file-operation driver hooks.
3106 stream->ops->poll_wait(stream, file, wait); in i915_perf_poll_locked()
3114 if (stream->pollin) in i915_perf_poll_locked()
3121 * i915_perf_poll - call poll_wait() with a suitable wait queue for stream
3135 struct i915_perf_stream *stream = file->private_data; in i915_perf_poll()
3136 struct i915_perf *perf = stream->perf; in i915_perf_poll()
3139 mutex_lock(&perf->lock); in i915_perf_poll()
3141 mutex_unlock(&perf->lock); in i915_perf_poll()
3147 * i915_perf_enable_locked - handle `I915_PERF_IOCTL_ENABLE` ioctl
3158 if (stream->enabled) in i915_perf_enable_locked()
3161 /* Allow stream->ops->enable() to refer to this */ in i915_perf_enable_locked()
3162 stream->enabled = true; in i915_perf_enable_locked()
3164 if (stream->ops->enable) in i915_perf_enable_locked()
3165 stream->ops->enable(stream); in i915_perf_enable_locked()
3167 if (stream->hold_preemption) in i915_perf_enable_locked()
3168 intel_context_set_nopreempt(stream->pinned_ctx); in i915_perf_enable_locked()
3172 * i915_perf_disable_locked - handle `I915_PERF_IOCTL_DISABLE` ioctl
3177 * The intention is that disabling an re-enabling a stream will ideally be
3178 * cheaper than destroying and re-opening a stream with the same configuration,
3180 * must be retained between disabling and re-enabling a stream.
3183 * to attempt to read from the stream (-EIO).
3187 if (!stream->enabled) in i915_perf_disable_locked()
3190 /* Allow stream->ops->disable() to refer to this */ in i915_perf_disable_locked()
3191 stream->enabled = false; in i915_perf_disable_locked()
3193 if (stream->hold_preemption) in i915_perf_disable_locked()
3194 intel_context_clear_nopreempt(stream->pinned_ctx); in i915_perf_disable_locked()
3196 if (stream->ops->disable) in i915_perf_disable_locked()
3197 stream->ops->disable(stream); in i915_perf_disable_locked()
3204 long ret = stream->oa_config->id; in i915_perf_config_locked()
3206 config = i915_perf_get_oa_config(stream->perf, metrics_set); in i915_perf_config_locked()
3208 return -EINVAL; in i915_perf_config_locked()
3210 if (config != stream->oa_config) { in i915_perf_config_locked()
3224 config = xchg(&stream->oa_config, config); in i915_perf_config_locked()
3235 * i915_perf_ioctl - support ioctl() usage with i915 perf stream FDs
3240 * Note: The &perf->lock mutex has been taken to serialize
3241 * with any non-file-operation driver hooks.
3243 * Returns: zero on success or a negative error code. Returns -EINVAL for
3261 return -EINVAL; in i915_perf_ioctl_locked()
3265 * i915_perf_ioctl - support ioctl() usage with i915 perf stream FDs
3272 * Returns: zero on success or a negative error code. Returns -EINVAL for
3279 struct i915_perf_stream *stream = file->private_data; in i915_perf_ioctl()
3280 struct i915_perf *perf = stream->perf; in i915_perf_ioctl()
3283 mutex_lock(&perf->lock); in i915_perf_ioctl()
3285 mutex_unlock(&perf->lock); in i915_perf_ioctl()
3291 * i915_perf_destroy_locked - destroy an i915 perf stream
3297 * Note: The &perf->lock mutex has been taken to serialize
3298 * with any non-file-operation driver hooks.
3302 if (stream->enabled) in i915_perf_destroy_locked()
3305 if (stream->ops->destroy) in i915_perf_destroy_locked()
3306 stream->ops->destroy(stream); in i915_perf_destroy_locked()
3308 if (stream->ctx) in i915_perf_destroy_locked()
3309 i915_gem_context_put(stream->ctx); in i915_perf_destroy_locked()
3315 * i915_perf_release - handles userspace close() of a stream file
3327 struct i915_perf_stream *stream = file->private_data; in i915_perf_release()
3328 struct i915_perf *perf = stream->perf; in i915_perf_release()
3330 mutex_lock(&perf->lock); in i915_perf_release()
3332 mutex_unlock(&perf->lock); in i915_perf_release()
3335 drm_dev_put(&perf->i915->drm); in i915_perf_release()
3356 * i915_perf_open_ioctl_locked - DRM ioctl() for userspace to open a stream FD
3365 * behalf of i915_perf_open_ioctl() with the &perf->lock mutex
3366 * taken to serialize with any non-file-operation driver hooks.
3392 if (props->single_context) { in i915_perf_open_ioctl_locked()
3393 u32 ctx_handle = props->ctx_handle; in i915_perf_open_ioctl_locked()
3394 struct drm_i915_file_private *file_priv = file->driver_priv; in i915_perf_open_ioctl_locked()
3400 ret = -ENOENT; in i915_perf_open_ioctl_locked()
3409 * non-privileged client. in i915_perf_open_ioctl_locked()
3411 * For Gen8->11 the OA unit no longer supports clock gating off for a in i915_perf_open_ioctl_locked()
3413 * from updating as system-wide / global values. Even though we can in i915_perf_open_ioctl_locked()
3424 if (IS_HASWELL(perf->i915) && specific_ctx) in i915_perf_open_ioctl_locked()
3426 else if (IS_GEN(perf->i915, 12) && specific_ctx && in i915_perf_open_ioctl_locked()
3427 (props->sample_flags & SAMPLE_OA_REPORT) == 0) in i915_perf_open_ioctl_locked()
3430 if (props->hold_preemption) { in i915_perf_open_ioctl_locked()
3431 if (!props->single_context) { in i915_perf_open_ioctl_locked()
3433 ret = -EINVAL; in i915_perf_open_ioctl_locked()
3442 if (props->has_sseu) in i915_perf_open_ioctl_locked()
3445 get_default_sseu_config(&props->sseu, props->engine); in i915_perf_open_ioctl_locked()
3455 ret = -EACCES; in i915_perf_open_ioctl_locked()
3461 ret = -ENOMEM; in i915_perf_open_ioctl_locked()
3465 stream->perf = perf; in i915_perf_open_ioctl_locked()
3466 stream->ctx = specific_ctx; in i915_perf_open_ioctl_locked()
3467 stream->poll_oa_period = props->poll_oa_period; in i915_perf_open_ioctl_locked()
3473 /* we avoid simply assigning stream->sample_flags = props->sample_flags in i915_perf_open_ioctl_locked()
3477 if (WARN_ON(stream->sample_flags != props->sample_flags)) { in i915_perf_open_ioctl_locked()
3478 ret = -ENODEV; in i915_perf_open_ioctl_locked()
3482 if (param->flags & I915_PERF_FLAG_FD_CLOEXEC) in i915_perf_open_ioctl_locked()
3484 if (param->flags & I915_PERF_FLAG_FD_NONBLOCK) in i915_perf_open_ioctl_locked()
3493 if (!(param->flags & I915_PERF_FLAG_DISABLED)) in i915_perf_open_ioctl_locked()
3499 drm_dev_get(&perf->i915->drm); in i915_perf_open_ioctl_locked()
3504 if (stream->ops->destroy) in i915_perf_open_ioctl_locked()
3505 stream->ops->destroy(stream); in i915_perf_open_ioctl_locked()
3517 return i915_cs_timestamp_ticks_to_ns(perf->i915, 2ULL << exponent); in oa_exponent_to_ns()
3521 * read_properties_unlocked - validate + copy userspace stream open properties
3545 props->poll_oa_period = DEFAULT_POLL_PERIOD_NS; in read_properties_unlocked()
3549 return -EINVAL; in read_properties_unlocked()
3552 /* At the moment we only support using i915-perf on the RCS. */ in read_properties_unlocked()
3553 props->engine = intel_engine_lookup_user(perf->i915, in read_properties_unlocked()
3556 if (!props->engine) { in read_properties_unlocked()
3557 DRM_DEBUG("No RENDER-capable engines\n"); in read_properties_unlocked()
3558 return -EINVAL; in read_properties_unlocked()
3569 return -EINVAL; in read_properties_unlocked()
3586 return -EINVAL; in read_properties_unlocked()
3591 props->single_context = 1; in read_properties_unlocked()
3592 props->ctx_handle = value; in read_properties_unlocked()
3596 props->sample_flags |= SAMPLE_OA_REPORT; in read_properties_unlocked()
3601 return -EINVAL; in read_properties_unlocked()
3603 props->metrics_set = value; in read_properties_unlocked()
3607 DRM_DEBUG("Out-of-range OA report format %llu\n", in read_properties_unlocked()
3609 return -EINVAL; in read_properties_unlocked()
3611 if (!perf->oa_formats[value].size) { in read_properties_unlocked()
3614 return -EINVAL; in read_properties_unlocked()
3616 props->oa_format = value; in read_properties_unlocked()
3622 return -EINVAL; in read_properties_unlocked()
3650 return -EACCES; in read_properties_unlocked()
3653 props->oa_periodic = true; in read_properties_unlocked()
3654 props->oa_period_exponent = value; in read_properties_unlocked()
3657 props->hold_preemption = !!value; in read_properties_unlocked()
3666 return -EFAULT; in read_properties_unlocked()
3669 ret = get_sseu_config(&props->sseu, props->engine, &user_sseu); in read_properties_unlocked()
3674 props->has_sseu = true; in read_properties_unlocked()
3681 return -EINVAL; in read_properties_unlocked()
3683 props->poll_oa_period = value; in read_properties_unlocked()
3687 return -EINVAL; in read_properties_unlocked()
3697 * i915_perf_open_ioctl - DRM ioctl() for userspace to open a stream FD
3707 * i915-perf stream is expected to be a suitable interface for other forms of
3714 * i915_perf_open_ioctl_locked() after taking the &perf->lock
3715 * mutex for serializing with any non-file-operation driver hooks.
3723 struct i915_perf *perf = &to_i915(dev)->perf; in i915_perf_open_ioctl()
3729 if (!perf->i915) { in i915_perf_open_ioctl()
3731 return -ENOTSUPP; in i915_perf_open_ioctl()
3737 if (param->flags & ~known_open_flags) { in i915_perf_open_ioctl()
3739 return -EINVAL; in i915_perf_open_ioctl()
3743 u64_to_user_ptr(param->properties_ptr), in i915_perf_open_ioctl()
3744 param->num_properties, in i915_perf_open_ioctl()
3749 mutex_lock(&perf->lock); in i915_perf_open_ioctl()
3751 mutex_unlock(&perf->lock); in i915_perf_open_ioctl()
3757 * i915_perf_register - exposes i915-perf to userspace
3762 * used to open an i915-perf stream.
3766 struct i915_perf *perf = &i915->perf; in i915_perf_register()
3768 if (!perf->i915) in i915_perf_register()
3775 mutex_lock(&perf->lock); in i915_perf_register()
3777 perf->metrics_kobj = in i915_perf_register()
3779 &i915->drm.primary->kdev->kobj); in i915_perf_register()
3781 mutex_unlock(&perf->lock); in i915_perf_register()
3785 * i915_perf_unregister - hide i915-perf from userspace
3788 * i915-perf state cleanup is split up into an 'unregister' and
3795 struct i915_perf *perf = &i915->perf; in i915_perf_unregister()
3797 if (!perf->metrics_kobj) in i915_perf_unregister()
3800 kobject_put(perf->metrics_kobj); in i915_perf_unregister()
3801 perf->metrics_kobj = NULL; in i915_perf_unregister()
3934 return ERR_PTR(-EINVAL); in alloc_oa_regs()
3938 return ERR_PTR(-ENOMEM); in alloc_oa_regs()
3949 err = -EINVAL; in alloc_oa_regs()
3977 return sprintf(buf, "%d\n", oa_config->id); in show_dynamic_id()
3983 sysfs_attr_init(&oa_config->sysfs_metric_id.attr); in create_dynamic_oa_sysfs_entry()
3984 oa_config->sysfs_metric_id.attr.name = "id"; in create_dynamic_oa_sysfs_entry()
3985 oa_config->sysfs_metric_id.attr.mode = S_IRUGO; in create_dynamic_oa_sysfs_entry()
3986 oa_config->sysfs_metric_id.show = show_dynamic_id; in create_dynamic_oa_sysfs_entry()
3987 oa_config->sysfs_metric_id.store = NULL; in create_dynamic_oa_sysfs_entry()
3989 oa_config->attrs[0] = &oa_config->sysfs_metric_id.attr; in create_dynamic_oa_sysfs_entry()
3990 oa_config->attrs[1] = NULL; in create_dynamic_oa_sysfs_entry()
3992 oa_config->sysfs_metric.name = oa_config->uuid; in create_dynamic_oa_sysfs_entry()
3993 oa_config->sysfs_metric.attrs = oa_config->attrs; in create_dynamic_oa_sysfs_entry()
3995 return sysfs_create_group(perf->metrics_kobj, in create_dynamic_oa_sysfs_entry()
3996 &oa_config->sysfs_metric); in create_dynamic_oa_sysfs_entry()
4000 * i915_perf_add_config_ioctl - DRM ioctl() for userspace to add a new OA config
4015 struct i915_perf *perf = &to_i915(dev)->perf; in i915_perf_add_config_ioctl()
4021 if (!perf->i915) { in i915_perf_add_config_ioctl()
4023 return -ENOTSUPP; in i915_perf_add_config_ioctl()
4026 if (!perf->metrics_kobj) { in i915_perf_add_config_ioctl()
4028 return -EINVAL; in i915_perf_add_config_ioctl()
4033 return -EACCES; in i915_perf_add_config_ioctl()
4036 if ((!args->mux_regs_ptr || !args->n_mux_regs) && in i915_perf_add_config_ioctl()
4037 (!args->boolean_regs_ptr || !args->n_boolean_regs) && in i915_perf_add_config_ioctl()
4038 (!args->flex_regs_ptr || !args->n_flex_regs)) { in i915_perf_add_config_ioctl()
4040 return -EINVAL; in i915_perf_add_config_ioctl()
4046 return -ENOMEM; in i915_perf_add_config_ioctl()
4049 oa_config->perf = perf; in i915_perf_add_config_ioctl()
4050 kref_init(&oa_config->ref); in i915_perf_add_config_ioctl()
4052 if (!uuid_is_valid(args->uuid)) { in i915_perf_add_config_ioctl()
4054 err = -EINVAL; in i915_perf_add_config_ioctl()
4058 /* Last character in oa_config->uuid will be 0 because oa_config is in i915_perf_add_config_ioctl()
4061 memcpy(oa_config->uuid, args->uuid, sizeof(args->uuid)); in i915_perf_add_config_ioctl()
4063 oa_config->mux_regs_len = args->n_mux_regs; in i915_perf_add_config_ioctl()
4065 perf->ops.is_valid_mux_reg, in i915_perf_add_config_ioctl()
4066 u64_to_user_ptr(args->mux_regs_ptr), in i915_perf_add_config_ioctl()
4067 args->n_mux_regs); in i915_perf_add_config_ioctl()
4074 oa_config->mux_regs = regs; in i915_perf_add_config_ioctl()
4076 oa_config->b_counter_regs_len = args->n_boolean_regs; in i915_perf_add_config_ioctl()
4078 perf->ops.is_valid_b_counter_reg, in i915_perf_add_config_ioctl()
4079 u64_to_user_ptr(args->boolean_regs_ptr), in i915_perf_add_config_ioctl()
4080 args->n_boolean_regs); in i915_perf_add_config_ioctl()
4087 oa_config->b_counter_regs = regs; in i915_perf_add_config_ioctl()
4089 if (INTEL_GEN(perf->i915) < 8) { in i915_perf_add_config_ioctl()
4090 if (args->n_flex_regs != 0) { in i915_perf_add_config_ioctl()
4091 err = -EINVAL; in i915_perf_add_config_ioctl()
4095 oa_config->flex_regs_len = args->n_flex_regs; in i915_perf_add_config_ioctl()
4097 perf->ops.is_valid_flex_reg, in i915_perf_add_config_ioctl()
4098 u64_to_user_ptr(args->flex_regs_ptr), in i915_perf_add_config_ioctl()
4099 args->n_flex_regs); in i915_perf_add_config_ioctl()
4106 oa_config->flex_regs = regs; in i915_perf_add_config_ioctl()
4109 err = mutex_lock_interruptible(&perf->metrics_lock); in i915_perf_add_config_ioctl()
4116 idr_for_each_entry(&perf->metrics_idr, tmp, id) { in i915_perf_add_config_ioctl()
4117 if (!strcmp(tmp->uuid, oa_config->uuid)) { in i915_perf_add_config_ioctl()
4119 err = -EADDRINUSE; in i915_perf_add_config_ioctl()
4131 oa_config->id = idr_alloc(&perf->metrics_idr, in i915_perf_add_config_ioctl()
4134 if (oa_config->id < 0) { in i915_perf_add_config_ioctl()
4136 err = oa_config->id; in i915_perf_add_config_ioctl()
4140 mutex_unlock(&perf->metrics_lock); in i915_perf_add_config_ioctl()
4142 DRM_DEBUG("Added config %s id=%i\n", oa_config->uuid, oa_config->id); in i915_perf_add_config_ioctl()
4144 return oa_config->id; in i915_perf_add_config_ioctl()
4147 mutex_unlock(&perf->metrics_lock); in i915_perf_add_config_ioctl()
4155 * i915_perf_remove_config_ioctl - DRM ioctl() for userspace to remove an OA config
4168 struct i915_perf *perf = &to_i915(dev)->perf; in i915_perf_remove_config_ioctl()
4173 if (!perf->i915) { in i915_perf_remove_config_ioctl()
4175 return -ENOTSUPP; in i915_perf_remove_config_ioctl()
4180 return -EACCES; in i915_perf_remove_config_ioctl()
4183 ret = mutex_lock_interruptible(&perf->metrics_lock); in i915_perf_remove_config_ioctl()
4187 oa_config = idr_find(&perf->metrics_idr, *arg); in i915_perf_remove_config_ioctl()
4190 ret = -ENOENT; in i915_perf_remove_config_ioctl()
4194 GEM_BUG_ON(*arg != oa_config->id); in i915_perf_remove_config_ioctl()
4196 sysfs_remove_group(perf->metrics_kobj, &oa_config->sysfs_metric); in i915_perf_remove_config_ioctl()
4198 idr_remove(&perf->metrics_idr, *arg); in i915_perf_remove_config_ioctl()
4200 mutex_unlock(&perf->metrics_lock); in i915_perf_remove_config_ioctl()
4202 DRM_DEBUG("Removed config %s id=%i\n", oa_config->uuid, oa_config->id); in i915_perf_remove_config_ioctl()
4209 mutex_unlock(&perf->metrics_lock); in i915_perf_remove_config_ioctl()
4256 * i915_perf_init - initialize i915-perf state on module bind
4259 * Initializes i915-perf state without exposing anything to userspace.
4261 * Note: i915-perf initialization is split into an 'init' and 'register'
4266 struct i915_perf *perf = &i915->perf; in i915_perf_init()
4271 perf->ops.is_valid_b_counter_reg = gen7_is_valid_b_counter_addr; in i915_perf_init()
4272 perf->ops.is_valid_mux_reg = hsw_is_valid_mux_addr; in i915_perf_init()
4273 perf->ops.is_valid_flex_reg = NULL; in i915_perf_init()
4274 perf->ops.enable_metric_set = hsw_enable_metric_set; in i915_perf_init()
4275 perf->ops.disable_metric_set = hsw_disable_metric_set; in i915_perf_init()
4276 perf->ops.oa_enable = gen7_oa_enable; in i915_perf_init()
4277 perf->ops.oa_disable = gen7_oa_disable; in i915_perf_init()
4278 perf->ops.read = gen7_oa_read; in i915_perf_init()
4279 perf->ops.oa_hw_tail_read = gen7_oa_hw_tail_read; in i915_perf_init()
4281 perf->oa_formats = hsw_oa_formats; in i915_perf_init()
4289 perf->ops.read = gen8_oa_read; in i915_perf_init()
4292 perf->oa_formats = gen8_plus_oa_formats; in i915_perf_init()
4294 perf->ops.is_valid_b_counter_reg = in i915_perf_init()
4296 perf->ops.is_valid_mux_reg = in i915_perf_init()
4298 perf->ops.is_valid_flex_reg = in i915_perf_init()
4302 perf->ops.is_valid_mux_reg = in i915_perf_init()
4306 perf->ops.oa_enable = gen8_oa_enable; in i915_perf_init()
4307 perf->ops.oa_disable = gen8_oa_disable; in i915_perf_init()
4308 perf->ops.enable_metric_set = gen8_enable_metric_set; in i915_perf_init()
4309 perf->ops.disable_metric_set = gen8_disable_metric_set; in i915_perf_init()
4310 perf->ops.oa_hw_tail_read = gen8_oa_hw_tail_read; in i915_perf_init()
4313 perf->ctx_oactxctrl_offset = 0x120; in i915_perf_init()
4314 perf->ctx_flexeu0_offset = 0x2ce; in i915_perf_init()
4316 perf->gen8_valid_ctx_bit = BIT(25); in i915_perf_init()
4318 perf->ctx_oactxctrl_offset = 0x128; in i915_perf_init()
4319 perf->ctx_flexeu0_offset = 0x3de; in i915_perf_init()
4321 perf->gen8_valid_ctx_bit = BIT(16); in i915_perf_init()
4324 perf->oa_formats = gen8_plus_oa_formats; in i915_perf_init()
4326 perf->ops.is_valid_b_counter_reg = in i915_perf_init()
4328 perf->ops.is_valid_mux_reg = in i915_perf_init()
4330 perf->ops.is_valid_flex_reg = in i915_perf_init()
4333 perf->ops.oa_enable = gen8_oa_enable; in i915_perf_init()
4334 perf->ops.oa_disable = gen8_oa_disable; in i915_perf_init()
4335 perf->ops.enable_metric_set = gen8_enable_metric_set; in i915_perf_init()
4336 perf->ops.disable_metric_set = gen10_disable_metric_set; in i915_perf_init()
4337 perf->ops.oa_hw_tail_read = gen8_oa_hw_tail_read; in i915_perf_init()
4340 perf->ctx_oactxctrl_offset = 0x128; in i915_perf_init()
4341 perf->ctx_flexeu0_offset = 0x3de; in i915_perf_init()
4343 perf->ctx_oactxctrl_offset = 0x124; in i915_perf_init()
4344 perf->ctx_flexeu0_offset = 0x78e; in i915_perf_init()
4346 perf->gen8_valid_ctx_bit = BIT(16); in i915_perf_init()
4348 perf->oa_formats = gen12_oa_formats; in i915_perf_init()
4350 perf->ops.is_valid_b_counter_reg = in i915_perf_init()
4352 perf->ops.is_valid_mux_reg = in i915_perf_init()
4354 perf->ops.is_valid_flex_reg = in i915_perf_init()
4357 perf->ops.oa_enable = gen12_oa_enable; in i915_perf_init()
4358 perf->ops.oa_disable = gen12_oa_disable; in i915_perf_init()
4359 perf->ops.enable_metric_set = gen12_enable_metric_set; in i915_perf_init()
4360 perf->ops.disable_metric_set = gen12_disable_metric_set; in i915_perf_init()
4361 perf->ops.oa_hw_tail_read = gen12_oa_hw_tail_read; in i915_perf_init()
4363 perf->ctx_flexeu0_offset = 0; in i915_perf_init()
4364 perf->ctx_oactxctrl_offset = 0x144; in i915_perf_init()
4368 if (perf->ops.enable_metric_set) { in i915_perf_init()
4369 mutex_init(&perf->lock); in i915_perf_init()
4372 RUNTIME_INFO(i915)->cs_timestamp_frequency_hz / 2; in i915_perf_init()
4374 mutex_init(&perf->metrics_lock); in i915_perf_init()
4375 idr_init(&perf->metrics_idr); in i915_perf_init()
4387 ratelimit_state_init(&perf->spurious_report_rs, 5 * HZ, 10); in i915_perf_init()
4392 ratelimit_set_flags(&perf->spurious_report_rs, in i915_perf_init()
4395 ratelimit_state_init(&perf->tail_pointer_race, in i915_perf_init()
4397 ratelimit_set_flags(&perf->tail_pointer_race, in i915_perf_init()
4400 atomic64_set(&perf->noa_programming_delay, in i915_perf_init()
4403 perf->i915 = i915; in i915_perf_init()
4424 * i915_perf_fini - Counter part to i915_perf_init()
4429 struct i915_perf *perf = &i915->perf; in i915_perf_fini()
4431 if (!perf->i915) in i915_perf_fini()
4434 idr_for_each(&perf->metrics_idr, destroy_config, perf); in i915_perf_fini()
4435 idr_destroy(&perf->metrics_idr); in i915_perf_fini()
4437 memset(&perf->ops, 0, sizeof(perf->ops)); in i915_perf_fini()
4438 perf->i915 = NULL; in i915_perf_fini()
4442 * i915_perf_ioctl_version - Version of the i915-perf subsystem